6.8.3.2.7. statsmodels.sandbox.stats.diagnostic.breaks_AP

statsmodels.sandbox.stats.diagnostic.breaks_AP(endog, exog, skip)

supLM, expLM and aveLM by Andrews, and Andrews,Ploberger

p-values by B Hansen

just idea for computation of sequence of tests with given change point (Chow tests) run recursive ols both forward and backward, match the two so they form a split of the data, calculate sum of squares for residuals and get test statistic for each breakpoint between skip and nobs-skip need to put recursive ols (residuals) into separate function

alternative: B Hansen loops over breakpoints only once and updates

x’x and xe’xe

update: Andrews is based on GMM estimation not OLS, LM test statistic is

easy to compute because it only requires full sample GMM estimate (p.837) with GMM the test has much wider applicability than just OLS

for testing loop over single breakpoint Chow test function